AI driven drug discovery pipeline for lead optimization: from virtual screening to preclinical validation of brain-penetrant ad therapeutics

The National Institute on Aging awarded the Trustees of Indiana University $1.232 million on February 1, 2026, under the Aging Research program (CFDA 93.866) to develop AI-driven virtual screening pipelines and computational tools for identifying blood-brain-barrier-penetrant drug candidates against Alzheimer's disease and related dementias.

The funded research encompasses three interconnected aims: developing AI-driven virtual screening pipelines incorporating small-molecule foundation models, Bayesian ensemble learning, and active learning to screen billion-scale compound libraries against AD/ADRD targets when docking-based structures are available; creating similarity-based virtual screening methods using affinity selection mass spectrometry and advanced AI tools for targets lacking ligand-bound structures; and constructing a large-scale dataset and developing machine-learning models to predict blood-brain barrier permeability metrics (Kp,uu and P-gp efflux ratios) for small-molecule drugs. The work includes up to 25 in vivo studies to benchmark permeability predictions and will produce open-source computational workflows and public databases.

Indiana Biosciences Research Institute, Inc. serves as a sub-recipient, performing laboratory work under all three aims including in vivo studies for pharmacokinetic exposure determination and blood-brain barrier penetration assessment, database development for quantum-chemical calculations, and publication and code release activities.

The award period runs from February 1, 2026, through January 31, 2031, with performance at Woodbridge, Indiana.

Abstract

In this proposed research, we aim to develop AI-driven virtual screening pipelines to screen billion-scale compound libraries and predict blood-brain barrier (BBB) permeability, which are critical steps for identifying drug-like lead compounds against ad/adrd targets, as the current computational methods do not adequately address these crucial needs. First, we will incorporate small molecule foundation models, bayesian ensemble learning, and active learning to accelerate the docking-based virtual screening pipeline, enabling efficient virtual screening of billion-scale compound libraries. Second, we will incorporate affinity selection mass spectrometry (ASMS)-based screening and advanced AI/computational tools to enable similarity-based virtual screening (SBVS) for ad/adrd targets when their ligand-bound structures are not known. Finally, we will construct a large- scale δgsolv dataset and take advantage of the dataset to develop AI/ML models that predict KP, UU and PGP ER, providing a more nuanced understanding of brain penetration for small-molecule brain penetrant drugs.